# Zed: The Open Source Agentic IDE - Use Claude Code, Codex & Gemini CLI in one place Source: https://www.youtube.com/watch?v=QU4hED-RZ5U Recap page: https://rapidrecap.app/video/QU4hED-RZ5U Generated: 2025-11-25T09:33:56.712+00:00 --- ## Quick Overview The Zed editor, an open-source agentic IDE, allows seamless integration and swapping between multiple AI agents like Claude Code, Codex CLI, and Gemini CLI directly within the editor interface, providing a unified environment for complex development tasks. **Key Points:** - Zed, an open-source agentic IDE, enables users to seamlessly integrate and switch between various AI agents, including Claude Code, Codex CLI, and Gemini CLI, using the Agent Client Protocol (ACP). - The demonstration shows a user initializing a new Next.js project and then interacting with the integrated AI agents to research topics like GPT-5.1 and Claude Sonnet 4.5. - The tool provides rich context awareness, showing plan execution steps (like using the `mcp_firecrawl_mcp_firecrawl_search` tool) and file changes during AI operations. - Users can control permission prompts for external tools, choosing between 'Always Ask', 'Accept Edits', 'Plan Mode', or 'Bypass Permissions' for different levels of automation. - The editor maintains a familiar look and feel compared to VS Code or Cursor, but offers the benefit of streaming token output and granular control over agent interactions. - The video highlights that this architecture allows for easy swapping between different models (e.g., Claude Sonnet 4 vs. Gemini 3) without major workflow disruption. ![Screenshot at 00:04: The Zed editor interface is shown with the agent selection dropdown open, displaying integrated agents like Claude Code, Codex CLI, and Gemini CLI, illustrating the core feature of agent switching.](https://ss.rapidrecap.app/screens/QU4hED-RZ5U/00-00-04.png) **Context:** The video demonstrates the capabilities of Zed, a next-generation, high-performance, open-source code editor built in Rust, focusing specifically on its agentic features that allow developers to integrate and coordinate different large language model (LLM) coding assistants via the Agent Client Protocol (ACP). The demonstration uses a newly created Next.js project to showcase how these agents can be summoned, execute complex multi-step plans involving web scraping (using an MCP tool), and provide feedback directly within the IDE environment. ## Detailed Analysis The video showcases Zed, an open-source, agentic IDE designed for high-performance collaboration, emphasizing its ability to integrate multiple AI agents using the Agent Client Protocol (ACP). The demonstration begins by setting up a new Next.js project via the terminal integrated within Zed. The user then interacts with the agent panel (top right) to select and switch between different AI agents, including Claude Code, Codex CLI, and Gemini CLI, demonstrating flexibility in model choice (e.g., Claude Sonnet 4). A complex task is initiated: researching the latest AI news regarding GPT-5.1, Gemini 3, and Anthropic's Sonnet 4.5. The agent breaks this down into a plan, executing external tools like `mcp_firecrawl_mcp_firecrawl_search` to gather information. The user observes the execution in real-time, including token streaming and step-by-step updates in the plan view. A key feature highlighted is the granular control over tool permissions, allowing users to select 'Always Ask', 'Accept Edits', 'Plan Mode', or 'Bypass Permissions' for specific actions, which is useful for high-trust or complex tasks. The demo concludes by showing the resulting high-quality blog post content generated by the AI, which incorporates the requested Neo-Brutalist design elements, proving the editor's capability to manage complex, multi-file, multi-agent workflows efficiently. ### Agent Integration and Switching - Agent selection menu visible at 00:01 - Seamless switching between Claude Code, Codex CLI, and Gemini CLI agents - Agents are managed via the Agent Client Protocol (ACP) ### Demonstration Workflow - Initializing a Next.js project using 'bunx create-next-app' at 4:08 - AI agent is tasked with researching AI models and creating a blog post - Agent executes search tools like 'mcp_firecrawl_mcp_firecrawl_search' to gather data ### Permission Control - Menu for setting agent permissions is shown at 7:27 - Options include 'Always Ask', 'Accept Edits', 'Plan Mode', and 'Bypass Permissions' - Bypass mode is noted as useful for low-stakes tasks but risky for large refactors ### AI Response Quality - Agent delivers a comprehensive plan and executes it iteratively (0:35, 6:40) - Generated content (a blog post on AI model wars) is displayed next to the code editor (11:50) ### Editor Experience - Zed editor features familiar VS Code/Cursor-like UI elements, including an integrated terminal and file explorer (3:44) - Performance metrics show fast compilation and rendering times (6:47) ![Screenshot at 00:01: Initial view of the Zed editor showing the file explorer, main editor pane, and the agent interaction sidebar.](https://ss.rapidrecap.app/screens/QU4hED-RZ5U/00-00-01.png) ![Screenshot at 00:16: The agent selection menu is open, displaying internal Zed agents and external agents like Claude Code, Codex CLI, and Gemini CLI.](https://ss.rapidrecap.app/screens/QU4hED-RZ5U/00-00-16.png) ![Screenshot at 00:24: The code editor pane \(tools.ts\) displaying TypeScript code related to tool extraction, alongside the agent chat panel showing file context awareness.](https://ss.rapidrecap.app/screens/QU4hED-RZ5U/00-00-24.png) ![Screenshot at 01:21: A macOS 'Force Quit Applications' window appears, indicating high memory usage by applications like Cursor, illustrating potential resource intensity when running multiple AI/dev tools.](https://ss.rapidrecap.app/screens/QU4hED-RZ5U/00-01-21.png) ![Screenshot at 05:54: The AI agent executes a multi-step plan, displaying the plan steps and tool execution results in the sidebar while the code editor is visible.](https://ss.rapidrecap.app/screens/QU4hED-RZ5U/00-05-54.png)